Re-enable CFS

kernel/src/lib.rs

@@ -37,10 +37,10 @@ use ostd::{
     cpu::{CpuId, CpuSet, PinCurrentCpu},
 };
 use process::Process;
+use sched::SchedPolicy;
 
 use crate::{
     prelude::*,
-    sched::priority::Priority,
     thread::{kernel_thread::ThreadOptions, Thread},
 };
 
@@ -86,8 +86,8 @@ pub fn main() {
     let mut affinity = CpuSet::new_empty();
     affinity.add(CpuId::bsp());
     ThreadOptions::new(init_thread)
-        .priority(Priority::idle())
         .cpu_affinity(affinity)
+        .sched_policy(SchedPolicy::Idle)
         .spawn();
 }
 
@@ -121,7 +121,7 @@ fn ap_init() {
 
     ThreadOptions::new(ap_idle_thread)
         .cpu_affinity(cpu_id.into())
-        .priority(Priority::idle())
+        .sched_policy(SchedPolicy::Idle)
         .spawn();
 }
 

kernel/src/net/iface/poll.rs

@@ -7,7 +7,11 @@ use log::trace;
 use ostd::timer::Jiffies;
 
 use super::{Iface, IFACES};
-use crate::{sched::priority::Priority, thread::kernel_thread::ThreadOptions, WaitTimeout};
+use crate::{
+    sched::{Nice, SchedPolicy},
+    thread::kernel_thread::ThreadOptions,
+    WaitTimeout,
+};
 
 pub fn lazy_init() {
     for iface in IFACES.get().unwrap() {
@@ -62,8 +66,7 @@ fn spawn_background_poll_thread(iface: Arc<Iface>) {
         }
     };
 
-    // FIXME: remove the use of real-time priority.
     ThreadOptions::new(task_fn)
-        .priority(Priority::default_real_time())
+        .sched_policy(SchedPolicy::Fair(Nice::MIN))
         .spawn();
 }

kernel/src/process/posix_thread/builder.rs

@@ -13,7 +13,7 @@ use crate::{
         signal::{sig_mask::AtomicSigMask, sig_queues::SigQueues},
         Credentials, Process,
     },
-    sched::priority::Priority,
+    sched::{Nice, SchedPolicy},
     thread::{task, Thread, Tid},
     time::{clocks::ProfClock, TimerManager},
 };
@@ -34,7 +34,7 @@ pub struct PosixThreadBuilder {
     fs: Option<Arc<ThreadFsInfo>>,
     sig_mask: AtomicSigMask,
     sig_queues: SigQueues,
-    priority: Priority,
+    sched_policy: SchedPolicy,
 }
 
 impl PosixThreadBuilder {
@@ -51,7 +51,7 @@ impl PosixThreadBuilder {
             fs: None,
             sig_mask: AtomicSigMask::new_empty(),
             sig_queues: SigQueues::new(),
-            priority: Priority::default(),
+            sched_policy: SchedPolicy::Fair(Nice::default()),
         }
     }
 
@@ -90,8 +90,8 @@ impl PosixThreadBuilder {
         self
     }
 
-    pub fn priority(mut self, priority: Priority) -> Self {
-        self.priority = priority;
+    pub fn sched_policy(mut self, sched_policy: SchedPolicy) -> Self {
+        self.sched_policy = sched_policy;
         self
     }
 
@@ -108,7 +108,7 @@ impl PosixThreadBuilder {
             fs,
             sig_mask,
             sig_queues,
-            priority,
+            sched_policy,
         } = self;
 
         let file_table = file_table.unwrap_or_else(|| RwArc::new(FileTable::new_with_stdio()));
@@ -141,8 +141,8 @@ impl PosixThreadBuilder {
             let thread = Arc::new(Thread::new(
                 weak_task.clone(),
                 posix_thread,
-                priority,
                 cpu_affinity,
+                sched_policy,
             ));
 
             let thread_local = ThreadLocal::new(set_child_tid, clear_child_tid, file_table);

kernel/src/process/process/builder.rs

@@ -12,7 +12,7 @@ use crate::{
         signal::sig_disposition::SigDispositions,
         Credentials,
     },
-    sched::priority::Nice,
+    sched::Nice,
 };
 
 pub struct ProcessBuilder<'a> {

kernel/src/process/process/mod.rs

@@ -20,7 +20,7 @@ use super::{
 use crate::{
     device::tty::open_ntty_as_controlling_terminal,
     prelude::*,
-    sched::priority::{AtomicNice, Nice},
+    sched::{AtomicNice, Nice},
     thread::{AsThread, Thread},
     time::clocks::ProfClock,
     vm::vmar::Vmar,

kernel/src/sched/mod.rs

@@ -1,16 +1,11 @@
 // SPDX-License-Identifier: MPL-2.0
 
-pub mod priority;
-// TODO: Remove this out-dated module once the `sched_class` module is stable.
-mod priority_scheduler;
+mod nice;
 mod sched_class;
 mod stats;
 
-// Export the stats getter functions.
-pub use stats::{loadavg, nr_queued_and_running};
-
-// TODO: Use `sched_class::init` instead after the completion of #1676.
-pub use self::priority_scheduler::init;
-// There may be multiple scheduling policies in the system,
-// and subsequent schedulers can be placed under this module.
-pub use self::sched_class::SchedAttr;
+pub use self::{
+    nice::{AtomicNice, Nice},
+    sched_class::{init, SchedAttr, SchedPolicy},
+    stats::{loadavg, nr_queued_and_running},
+};

kernel/src/sched/nice.rs

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: MPL-2.0
 
-use core::sync::atomic::{AtomicI8, AtomicU8};
+use core::sync::atomic::AtomicI8;
 
 use atomic_integer_wrapper::define_atomic_version_of_integer_like_type;
 
@@ -9,9 +9,9 @@ use atomic_integer_wrapper::define_atomic_version_of_integer_like_type;
 /// It is an integer in the range of [-20, 19]. Process with a smaller nice
 /// value is more favorable in scheduling.
 #[derive(Clone, Copy, Debug, Eq, PartialEq, PartialOrd, Ord)]
-pub struct Nice(NiceRange);
+pub struct Nice(NiceValue);
 
-pub type NiceRange = RangedI8<-20, 19>;
+pub type NiceValue = RangedI8<-20, 19>;
 
 define_atomic_version_of_integer_like_type!(Nice, try_from = true, {
     #[derive(Debug)]
@@ -19,22 +19,25 @@ define_atomic_version_of_integer_like_type!(Nice, try_from = true, {
 });
 
 impl Nice {
-    pub const fn new(range: NiceRange) -> Self {
+    pub const MIN: Self = Nice::new(NiceValue::MIN);
+    pub const MAX: Self = Nice::new(NiceValue::MAX);
+
+    pub const fn new(range: NiceValue) -> Self {
         Self(range)
     }
 
-    pub const fn range(&self) -> &NiceRange {
+    pub const fn value(&self) -> &NiceValue {
         &self.0
     }
 
-    pub fn range_mut(&mut self) -> &mut NiceRange {
+    pub fn value_mut(&mut self) -> &mut NiceValue {
         &mut self.0
     }
 }
 
 impl Default for Nice {
     fn default() -> Self {
-        Self::new(NiceRange::new(0))
+        Self::new(NiceValue::new(0))
     }
 }
 
@@ -45,7 +48,7 @@ impl From<Nice> for i8 {
 }
 
 impl TryFrom<i8> for Nice {
-    type Error = <NiceRange as TryFrom<i8>>::Error;
+    type Error = <NiceValue as TryFrom<i8>>::Error;
 
     fn try_from(value: i8) -> Result<Self, Self::Error> {
         let range = value.try_into()?;
@@ -53,79 +56,14 @@ impl TryFrom<i8> for Nice {
     }
 }
 
-/// The thread scheduling priority value.
-///
-/// It is an integer in the range of [0, 139]. Here we follow the Linux
-/// priority mappings: the relation between [`Priority`] and [`Nice`] is
-/// as such - prio = nice + 120 while the priority of [0, 99] are
-/// reserved for real-time tasks.
-#[derive(Clone, Copy, Debug, Eq, PartialEq, PartialOrd, Ord)]
-pub struct Priority(PriorityRange);
-
-pub type PriorityRange = RangedU8<0, 139>;
-
-define_atomic_version_of_integer_like_type!(Priority, try_from = true, {
-    #[derive(Debug)]
-    pub struct AtomicPriority(AtomicU8);
-});
-
-impl Priority {
-    pub const MIN_NORMAL: Self = Self::new(PriorityRange::new(100));
-
-    pub const fn new(range: PriorityRange) -> Self {
-        Self(range)
-    }
-
-    pub const fn range(&self) -> &PriorityRange {
-        &self.0
-    }
-
-    pub fn range_mut(&mut self) -> &mut PriorityRange {
-        &mut self.0
-    }
-}
-
-impl From<Nice> for Priority {
-    fn from(nice: Nice) -> Self {
-        Self::new(PriorityRange::new((nice.range().get() as i16 + 120) as u8))
-    }
-}
-
-impl From<Priority> for Nice {
-    fn from(priority: Priority) -> Self {
-        Self::new(NiceRange::new(((priority.range().get() as i16 - 100) - 20) as i8))
-    }
-}
-
-impl Default for Priority {
-    fn default() -> Self {
-        Nice::default().into()
-    }
-}
-
-impl From<Priority> for u8 {
-    fn from(value: Priority) -> Self {
-        value.0.into()
-    }
-}
-
-impl TryFrom<u8> for Priority {
-    type Error = <PriorityRange as TryFrom<u8>>::Error;
-
-    fn try_from(value: u8) -> Result<Self, Self::Error> {
-        let range = value.try_into()?;
-        Ok(Self::new(range))
-    }
-}
-
 macro_rules! define_ranged_integer {
     ($visibility: vis, $name: ident, $type: ty) => {
         #[derive(Copy, Clone, Debug, Eq, PartialEq, PartialOrd, Ord)]
         $visibility struct $name<const MIN: $type, const MAX: $type>($type);
 
         impl<const MIN: $type, const MAX: $type> $name<MIN, MAX> {
-            $visibility const MIN: $type = MIN as $type;
-            $visibility const MAX: $type = MAX as $type;
+            $visibility const MIN: Self = Self::new(MIN);
+            $visibility const MAX: Self = Self::new(MAX);
 
             $visibility const fn new(val: $type) -> Self {
                 assert!(val >= MIN && val <= MAX);
@@ -152,7 +90,7 @@ macro_rules! define_ranged_integer {
             type Error = &'static str;
 
             fn try_from(value: $type) -> Result<Self, Self::Error> {
-                if value < Self::MIN || value > Self::MAX {
+                if value < MIN || value > MAX {
                     Err("Initialized with out-of-range value.")
                 } else {
                     Ok(Self(value))

kernel/src/sched/priority_scheduler.rs

@@ -1,338 +0,0 @@
-// SPDX-License-Identifier: MPL-2.0
-
-use core::sync::atomic::Ordering;
-
-use ostd::{
-    cpu::{num_cpus, CpuId, CpuSet, PinCurrentCpu},
-    task::{
-        disable_preempt,
-        scheduler::{
-            info::CommonSchedInfo, inject_scheduler, EnqueueFlags, LocalRunQueue, Scheduler,
-            UpdateFlags,
-        },
-        Task,
-    },
-    trap::disable_local,
-};
-
-use super::{
-    priority::Priority,
-    stats::{set_stats_from_scheduler, SchedulerStats},
-};
-use crate::{prelude::*, thread::Thread};
-
-pub fn init() {
-    let preempt_scheduler = Box::new(PreemptScheduler::default());
-    let scheduler = Box::<PreemptScheduler<Thread, Task>>::leak(preempt_scheduler);
-
-    // Inject the scheduler into the ostd for actual scheduling work.
-    inject_scheduler(scheduler);
-
-    // Set the scheduler into the system for statistics.
-    // We set this after injecting the scheduler into ostd,
-    // so that the loadavg statistics are updated after the scheduler is used.
-    set_stats_from_scheduler(scheduler);
-}
-
-/// The preempt scheduler.
-///
-/// Real-time tasks are placed in the `real_time_entities` queue and
-/// are always prioritized during scheduling.
-/// Normal tasks are placed in the `normal_entities` queue and are only
-/// scheduled for execution when there are no real-time tasks.
-struct PreemptScheduler<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> {
-    rq: Vec<SpinLock<PreemptRunQueue<T, U>>>,
-}
-
-impl<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> PreemptScheduler<T, U> {
-    fn new(nr_cpus: usize) -> Self {
-        let mut rq = Vec::with_capacity(nr_cpus);
-        for _ in 0..nr_cpus {
-            rq.push(SpinLock::new(PreemptRunQueue::new()));
-        }
-        Self { rq }
-    }
-
-    /// Selects a CPU for task to run on for the first time.
-    fn select_cpu(&self, entity: &PreemptSchedEntity<T, U>) -> CpuId {
-        // If the CPU of a runnable task has been set before, keep scheduling
-        // the task to that one.
-        // TODO: Consider migrating tasks between CPUs for load balancing.
-        if let Some(cpu_id) = entity.task.cpu().get() {
-            return cpu_id;
-        }
-
-        let irq_guard = disable_local();
-        let mut selected = irq_guard.current_cpu();
-        let mut minimum_load = usize::MAX;
-
-        for candidate in entity.thread.cpu_affinity().iter() {
-            let rq = self.rq[candidate.as_usize()].lock();
-            // A wild guess measuring the load of a runqueue. We assume that
-            // real-time tasks are 4-times as important as normal tasks.
-            let load = rq.real_time_entities.len() * 8
-                + rq.normal_entities.len() * 2
-                + rq.lowest_entities.len();
-            if load < minimum_load {
-                selected = candidate;
-                minimum_load = load;
-            }
-        }
-
-        selected
-    }
-}
-
-impl<T: Sync + Send + PreemptSchedInfo + FromTask<U>, U: Sync + Send + CommonSchedInfo> Scheduler<U>
-    for PreemptScheduler<T, U>
-{
-    fn enqueue(&self, task: Arc<U>, flags: EnqueueFlags) -> Option<CpuId> {
-        let entity = PreemptSchedEntity::new(task);
-
-        let (still_in_rq, target_cpu) = {
-            let selected_cpu_id = self.select_cpu(&entity);
-
-            if let Err(task_cpu_id) = entity.task.cpu().set_if_is_none(selected_cpu_id) {
-                debug_assert!(flags != EnqueueFlags::Spawn);
-                (true, task_cpu_id)
-            } else {
-                (false, selected_cpu_id)
-            }
-        };
-
-        let mut rq = self.rq[target_cpu.as_usize()].disable_irq().lock();
-        if still_in_rq && let Err(_) = entity.task.cpu().set_if_is_none(target_cpu) {
-            return None;
-        }
-
-        let new_priority = entity.thread.priority();
-
-        if entity.thread.is_real_time() {
-            rq.real_time_entities.push_back(entity);
-        } else if entity.thread.is_lowest() {
-            rq.lowest_entities.push_back(entity);
-        } else {
-            rq.normal_entities.push_back(entity);
-        }
-
-        // Preempt the current task, but only if the newly queued task has a strictly higher
-        // priority (i.e., a lower value returned by the `priority` method) than the current task.
-        if rq
-            .current
-            .as_ref()
-            .is_some_and(|current| new_priority < current.thread.priority())
-        {
-            Some(target_cpu)
-        } else {
-            None
-        }
-    }
-
-    fn local_rq_with(&self, f: &mut dyn FnMut(&dyn LocalRunQueue<U>)) {
-        let irq_guard = disable_local();
-        let local_rq: &PreemptRunQueue<T, U> = &self.rq[irq_guard.current_cpu().as_usize()].lock();
-        f(local_rq);
-    }
-
-    fn local_mut_rq_with(&self, f: &mut dyn FnMut(&mut dyn LocalRunQueue<U>)) {
-        let irq_guard = disable_local();
-        let local_rq: &mut PreemptRunQueue<T, U> =
-            &mut self.rq[irq_guard.current_cpu().as_usize()].lock();
-        f(local_rq);
-    }
-}
-
-impl<T: Sync + Send + PreemptSchedInfo + FromTask<U>, U: Sync + Send + CommonSchedInfo>
-    SchedulerStats for PreemptScheduler<T, U>
-{
-    fn nr_queued_and_running(&self) -> (u32, u32) {
-        let _preempt_guard = disable_preempt();
-        let mut nr_queued = 0;
-        let mut nr_running = 0;
-
-        for rq in self.rq.iter() {
-            let rq = rq.lock();
-
-            nr_queued +=
-                rq.real_time_entities.len() + rq.normal_entities.len() + rq.lowest_entities.len();
-
-            if rq.current.is_some() {
-                nr_running += 1;
-            }
-        }
-
-        (nr_queued as u32, nr_running)
-    }
-}
-
-impl Default for PreemptScheduler<Thread, Task> {
-    fn default() -> Self {
-        Self::new(num_cpus())
-    }
-}
-
-struct PreemptRunQueue<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> {
-    current: Option<PreemptSchedEntity<T, U>>,
-    real_time_entities: VecDeque<PreemptSchedEntity<T, U>>,
-    normal_entities: VecDeque<PreemptSchedEntity<T, U>>,
-    lowest_entities: VecDeque<PreemptSchedEntity<T, U>>,
-}
-
-impl<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> PreemptRunQueue<T, U> {
-    pub fn new() -> Self {
-        Self {
-            current: None,
-            real_time_entities: VecDeque::new(),
-            normal_entities: VecDeque::new(),
-            lowest_entities: VecDeque::new(),
-        }
-    }
-}
-
-impl<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> LocalRunQueue<U>
-    for PreemptRunQueue<T, U>
-{
-    fn current(&self) -> Option<&Arc<U>> {
-        self.current.as_ref().map(|entity| &entity.task)
-    }
-
-    fn update_current(&mut self, flags: UpdateFlags) -> bool {
-        match flags {
-            UpdateFlags::Tick => {
-                let Some(ref mut current_entity) = self.current else {
-                    return false;
-                };
-                current_entity.tick()
-                    || (!current_entity.thread.is_real_time()
-                        && !self.real_time_entities.is_empty())
-            }
-            _ => true,
-        }
-    }
-
-    fn pick_next_current(&mut self) -> Option<&Arc<U>> {
-        let next_entity = if !self.real_time_entities.is_empty() {
-            self.real_time_entities.pop_front()
-        } else if !self.normal_entities.is_empty() {
-            self.normal_entities.pop_front()
-        } else {
-            self.lowest_entities.pop_front()
-        }?;
-        if let Some(prev_entity) = self.current.replace(next_entity) {
-            if prev_entity.thread.is_real_time() {
-                self.real_time_entities.push_back(prev_entity);
-            } else if prev_entity.thread.is_lowest() {
-                self.lowest_entities.push_back(prev_entity);
-            } else {
-                self.normal_entities.push_back(prev_entity);
-            }
-        }
-
-        Some(&self.current.as_ref().unwrap().task)
-    }
-
-    fn dequeue_current(&mut self) -> Option<Arc<U>> {
-        self.current.take().map(|entity| {
-            let runnable = entity.task;
-            runnable.cpu().set_to_none();
-
-            runnable
-        })
-    }
-}
-struct PreemptSchedEntity<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> {
-    task: Arc<U>,
-    thread: Arc<T>,
-    time_slice: TimeSlice,
-}
-
-impl<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> PreemptSchedEntity<T, U> {
-    fn new(task: Arc<U>) -> Self {
-        let thread = T::from_task(&task);
-        let time_slice = TimeSlice::default();
-        Self {
-            task,
-            thread,
-            time_slice,
-        }
-    }
-
-    fn tick(&mut self) -> bool {
-        self.time_slice.elapse()
-    }
-}
-
-impl<T: PreemptSchedInfo + FromTask<U>, U: CommonSchedInfo> Clone for PreemptSchedEntity<T, U> {
-    fn clone(&self) -> Self {
-        Self {
-            task: self.task.clone(),
-            thread: self.thread.clone(),
-            time_slice: self.time_slice,
-        }
-    }
-}
-
-#[derive(Clone, Copy)]
-pub struct TimeSlice {
-    elapsed_ticks: u32,
-}
-
-impl TimeSlice {
-    const DEFAULT_TIME_SLICE: u32 = 100;
-
-    pub const fn new() -> Self {
-        TimeSlice { elapsed_ticks: 0 }
-    }
-
-    pub fn elapse(&mut self) -> bool {
-        self.elapsed_ticks = (self.elapsed_ticks + 1) % Self::DEFAULT_TIME_SLICE;
-
-        self.elapsed_ticks == 0
-    }
-}
-
-impl Default for TimeSlice {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
-impl PreemptSchedInfo for Thread {
-    const REAL_TIME_TASK_PRIORITY: Priority = Priority::default_real_time();
-    const LOWEST_TASK_PRIORITY: Priority = Priority::idle();
-
-    fn priority(&self) -> Priority {
-        self.atomic_priority().load(Ordering::Relaxed)
-    }
-
-    fn cpu_affinity(&self) -> CpuSet {
-        self.atomic_cpu_affinity().load()
-    }
-}
-
-trait PreemptSchedInfo {
-    const REAL_TIME_TASK_PRIORITY: Priority;
-    const LOWEST_TASK_PRIORITY: Priority;
-
-    fn priority(&self) -> Priority;
-
-    fn cpu_affinity(&self) -> CpuSet;
-
-    fn is_real_time(&self) -> bool {
-        self.priority() < Self::REAL_TIME_TASK_PRIORITY
-    }
-
-    fn is_lowest(&self) -> bool {
-        self.priority() == Self::LOWEST_TASK_PRIORITY
-    }
-}
-
-impl FromTask<Task> for Thread {
-    fn from_task(task: &Arc<Task>) -> Arc<Self> {
-        task.data().downcast_ref::<Arc<Self>>().unwrap().clone()
-    }
-}
-
-trait FromTask<U> {
-    fn from_task(task: &Arc<U>) -> Arc<Self>;
-}

kernel/src/sched/sched_class/fair.rs

@@ -19,7 +19,7 @@ use super::{
     CurrentRuntime, SchedAttr, SchedClassRq,
 };
 use crate::{
-    sched::priority::{Nice, NiceRange},
+    sched::nice::{Nice, NiceValue},
     thread::AsThread,
 };
 
@@ -38,8 +38,8 @@ pub const fn nice_to_weight(nice: Nice) -> u64 {
         let mut ret = [0; 40];
 
         let mut index = 0;
-        let mut nice = NiceRange::MIN;
-        while nice <= NiceRange::MAX {
+        let mut nice = NiceValue::MIN.get();
+        while nice <= NiceValue::MAX.get() {
             ret[index] = match nice {
                 0 => WEIGHT_0,
                 nice @ 1.. => {
@@ -60,7 +60,7 @@ pub const fn nice_to_weight(nice: Nice) -> u64 {
         ret
     };
 
-    NICE_TO_WEIGHT[(nice.range().get() + 20) as usize]
+    NICE_TO_WEIGHT[(nice.value().get() + 20) as usize]
 }
 
 /// The scheduling entity for the FAIR scheduling class.

kernel/src/sched/sched_class/mod.rs

@@ -19,7 +19,10 @@ use ostd::{
     trap::disable_local,
 };
 
-use super::{priority::Nice, stats::SchedulerStats};
+use super::{
+    nice::Nice,
+    stats::{set_stats_from_scheduler, SchedulerStats},
+};
 use crate::thread::{AsThread, Thread};
 
 mod policy;
@@ -30,13 +33,21 @@ mod idle;
 mod real_time;
 mod stop;
 
-use self::policy::{SchedPolicy, SchedPolicyKind, SchedPolicyState};
+pub use self::policy::SchedPolicy;
+use self::policy::{SchedPolicyKind, SchedPolicyState};
 
 type SchedEntity = (Arc<Task>, Arc<Thread>);
 
-#[expect(unused)]
 pub fn init() {
-    inject_scheduler(Box::leak(Box::new(ClassScheduler::new())));
+    let scheduler = Box::leak(Box::new(ClassScheduler::new()));
+
+    // Inject the scheduler into the ostd for actual scheduling work.
+    inject_scheduler(scheduler);
+
+    // Set the scheduler into the system for statistics.
+    // We set this after injecting the scheduler into ostd,
+    // so that the loadavg statistics are updated after the scheduler is used.
+    set_stats_from_scheduler(scheduler);
 }
 
 /// Represents the middle layer between scheduling classes and generic scheduler
@@ -128,7 +139,7 @@ impl SchedAttr {
             real_time: {
                 let (prio, policy) = match policy {
                     SchedPolicy::RealTime { rt_prio, rt_policy } => (rt_prio.get(), rt_policy),
-                    _ => (real_time::RtPrio::MAX, Default::default()),
+                    _ => (real_time::RtPrio::MAX.get(), Default::default()),
                 };
                 real_time::RealTimeAttr::new(prio, policy)
             },

kernel/src/sched/sched_class/policy.rs

@@ -7,7 +7,7 @@ use int_to_c_enum::TryFromInt;
 use ostd::sync::SpinLock;
 
 pub use super::real_time::{RealTimePolicy, RtPrio};
-use crate::sched::priority::{Nice, Priority};
+use crate::sched::nice::Nice;
 
 /// The User-chosen scheduling policy.
 ///
@@ -16,7 +16,7 @@ use crate::sched::priority::{Nice, Priority};
 pub enum SchedPolicy {
     Stop,
     RealTime {
-        rt_prio: super::real_time::RtPrio,
+        rt_prio: RtPrio,
         rt_policy: RealTimePolicy,
     },
     Fair(Nice),
@@ -32,19 +32,6 @@ pub(super) enum SchedPolicyKind {
     Idle = 3,
 }
 
-impl From<Priority> for SchedPolicy {
-    fn from(priority: Priority) -> Self {
-        match priority.range().get() {
-            rt @ 0..=99 => SchedPolicy::RealTime {
-                rt_prio: RtPrio::new(rt as u8),
-                rt_policy: Default::default(),
-            },
-            100..=139 => SchedPolicy::Fair(priority.into()),
-            _ => unreachable!(),
-        }
-    }
-}
-
 impl SchedPolicy {
     pub(super) fn kind(&self) -> SchedPolicyKind {
         match self {

kernel/src/sched/sched_class/real_time.rs

@@ -17,7 +17,7 @@ use ostd::{
 };
 
 use super::{time::base_slice_clocks, CurrentRuntime, SchedAttr, SchedClassRq};
-use crate::{sched::priority::RangedU8, thread::AsThread};
+use crate::{sched::nice::RangedU8, thread::AsThread};
 
 pub type RtPrio = RangedU8<0, 99>;
 

kernel/src/syscall/set_get_priority.rs

@@ -6,14 +6,17 @@ use super::SyscallReturn;
 use crate::{
     prelude::*,
     process::{posix_thread::AsPosixThread, process_table, Pgid, Pid, Process, Uid},
-    sched::priority::{Nice, NiceRange},
+    sched::Nice,
 };
 
 pub fn sys_set_priority(which: i32, who: u32, prio: i32, ctx: &Context) -> Result<SyscallReturn> {
     let prio_target = PriorityTarget::new(which, who, ctx)?;
-    let new_nice = {
-        let nice_raw = prio.clamp(NiceRange::MIN as i32, NiceRange::MAX as i32) as i8;
-        Nice::new(NiceRange::new(nice_raw))
+    let new_nice: Nice = {
+        let nice_raw = prio.clamp(
+            Nice::MIN.value().get() as i32,
+            Nice::MAX.value().get() as i32,
+        ) as i8;
+        nice_raw.try_into().unwrap()
     };
 
     debug!(
@@ -35,9 +38,9 @@ pub fn sys_get_priority(which: i32, who: u32, ctx: &Context) -> Result<SyscallRe
 
     let processes = get_processes(prio_target)?;
     let highest_prio = {
-        let mut nice = NiceRange::MAX;
+        let mut nice = Nice::MAX.value().get();
         for process in processes.iter() {
-            let proc_nice = process.nice().load(Ordering::Relaxed).range().get();
+            let proc_nice = process.nice().load(Ordering::Relaxed).value().get();
             // Returns the highest priority enjoyed by the processes
             if proc_nice < nice {
                 nice = proc_nice;

kernel/src/thread/kernel_thread.rs

@@ -6,7 +6,10 @@ use ostd::{
 };
 
 use super::{oops, AsThread, Thread};
-use crate::{prelude::*, sched::priority::Priority};
+use crate::{
+    prelude::*,
+    sched::{Nice, SchedPolicy},
+};
 
 /// The inner data of a kernel thread.
 struct KernelThread;
@@ -14,8 +17,8 @@ struct KernelThread;
 /// Options to create or spawn a new kernel thread.
 pub struct ThreadOptions {
     func: Option<Box<dyn FnOnce() + Send>>,
-    priority: Priority,
     cpu_affinity: CpuSet,
+    sched_policy: SchedPolicy,
 }
 
 impl ThreadOptions {
@@ -25,24 +28,25 @@ impl ThreadOptions {
         F: FnOnce() + Send + 'static,
     {
         let cpu_affinity = CpuSet::new_full();
+        let sched_policy = SchedPolicy::Fair(Nice::default());
         Self {
             func: Some(Box::new(func)),
-            priority: Priority::default(),
             cpu_affinity,
+            sched_policy,
         }
     }
 
-    /// Sets the priority of the new thread.
-    pub fn priority(mut self, priority: Priority) -> Self {
-        self.priority = priority;
-        self
-    }
-
     /// Sets the CPU affinity of the new thread.
     pub fn cpu_affinity(mut self, cpu_affinity: CpuSet) -> Self {
         self.cpu_affinity = cpu_affinity;
         self
     }
+
+    /// Sets the scheduling policy.
+    pub fn sched_policy(mut self, sched_policy: SchedPolicy) -> Self {
+        self.sched_policy = sched_policy;
+        self
+    }
 }
 
 impl ThreadOptions {
@@ -58,13 +62,13 @@ impl ThreadOptions {
         Arc::new_cyclic(|weak_task| {
             let thread = {
                 let kernel_thread = KernelThread;
-                let priority = self.priority;
                 let cpu_affinity = self.cpu_affinity;
+                let sched_policy = self.sched_policy;
                 Arc::new(Thread::new(
                     weak_task.clone(),
                     kernel_thread,
-                    priority,
                     cpu_affinity,
+                    sched_policy,
                 ))
             };
 

kernel/src/thread/mod.rs

@@ -12,10 +12,7 @@ use ostd::{
 use self::status::{AtomicThreadStatus, ThreadStatus};
 use crate::{
     prelude::*,
-    sched::{
-        priority::{AtomicPriority, Priority},
-        SchedAttr,
-    },
+    sched::{SchedAttr, SchedPolicy},
 };
 
 pub mod exception;
@@ -39,8 +36,6 @@ pub struct Thread {
     // mutable part
     /// Thread status
     status: AtomicThreadStatus,
-    /// Thread priority
-    priority: AtomicPriority,
     /// Thread CPU affinity
     cpu_affinity: AtomicCpuSet,
     sched_attr: SchedAttr,
@@ -51,16 +46,15 @@ impl Thread {
     pub fn new(
         task: Weak<Task>,
         data: impl Send + Sync + Any,
-        priority: Priority,
         cpu_affinity: CpuSet,
+        sched_policy: SchedPolicy,
     ) -> Self {
         Thread {
             task,
             data: Box::new(data),
             status: AtomicThreadStatus::new(ThreadStatus::Init),
-            priority: AtomicPriority::new(priority),
             cpu_affinity: AtomicCpuSet::new(cpu_affinity),
-            sched_attr: SchedAttr::new(priority.into()),
+            sched_attr: SchedAttr::new(sched_policy),
         }
     }
 
@@ -135,11 +129,6 @@ impl Thread {
         self.status.store(ThreadStatus::Exited, Ordering::Release);
     }
 
-    /// Returns the reference to the atomic priority.
-    pub fn atomic_priority(&self) -> &AtomicPriority {
-        &self.priority
-    }
-
     /// Returns the reference to the atomic CPU affinity.
     pub fn atomic_cpu_affinity(&self) -> &AtomicCpuSet {
         &self.cpu_affinity

kernel/src/thread/work_queue/worker.rs

@@ -10,7 +10,7 @@ use ostd::{
 use super::worker_pool::WorkerPool;
 use crate::{
     prelude::*,
-    sched::priority::Priority,
+    sched::{Nice, SchedPolicy},
     thread::{kernel_thread::ThreadOptions, AsThread},
 };
 
@@ -50,13 +50,15 @@ impl Worker {
             });
             let mut cpu_affinity = CpuSet::new_empty();
             cpu_affinity.add(bound_cpu);
-            let mut priority = Priority::default();
-            if worker_pool.upgrade().unwrap().is_high_priority() {
-                priority = Priority::default_real_time();
-            }
+            let sched_policy =
+                SchedPolicy::Fair(if worker_pool.upgrade().unwrap().is_high_priority() {
+                    Nice::MIN
+                } else {
+                    Nice::default()
+                });
             let bound_task = ThreadOptions::new(task_fn)
                 .cpu_affinity(cpu_affinity)
-                .priority(priority)
+                .sched_policy(sched_policy)
                 .build();
             Self {
                 worker_pool,

kernel/src/thread/work_queue/worker_pool.rs

@@ -16,7 +16,7 @@ use ostd::{
 use super::{simple_scheduler::SimpleScheduler, worker::Worker, WorkItem, WorkPriority, WorkQueue};
 use crate::{
     prelude::*,
-    sched::priority::Priority,
+    sched::{Nice, SchedPolicy},
     thread::{kernel_thread::ThreadOptions, AsThread},
 };
 
@@ -237,17 +237,13 @@ impl Monitor {
                 current_monitor.run_monitor_loop();
             });
             let cpu_affinity = CpuSet::new_full();
-            // FIXME: remove the use of real-time priority.
-            // Logically all monitors should be of default normal priority.
-            // This workaround is to make the monitor of high-priority worker pool
-            // starvation-free under the current scheduling policy.
-            let priority = match priority {
-                WorkPriority::High => Priority::default_real_time(),
-                WorkPriority::Normal => Priority::default(),
-            };
+            let sched_policy = SchedPolicy::Fair(match priority {
+                WorkPriority::High => Nice::MIN,
+                WorkPriority::Normal => Nice::default(),
+            });
             let bound_task = ThreadOptions::new(task_fn)
                 .cpu_affinity(cpu_affinity)
-                .priority(priority)
+                .sched_policy(sched_policy)
                 .build();
             Self {
                 worker_pool,